Terence Tao wants some mathematical problems kept off limits to AI solvers
热度趋势
趋势数据积累中
百分比基于当前可用热度信号，而非评论数或独立用户人数。
数学家陶哲轩（Terence Tao）建议，一些数学问题应避免由人工智能（AI）解决。他认为，在AI出现之前存在的开放性问题是有限的“未受污染”的基准。一旦某个问题的解决方案被公布，就很难判断未来的AI是独立解决了该问题，还是在训练过程中接触过答案。陶哲轩还认为，开放性问题对于培养数学家和发展新方法具有重要价值。因此，他建议数学界最终可以通过社会规范，将某些问题指定为自动化求解器的禁区，转而引导AI解决其他问题。
tl;dr: Tao’s argument is that pre-AI open problems are now a limited supply of “uncontaminated” benchmarks. Once someone publishes a solution, you can’t easily tell whether a future AI independently solved it or had access to the answer during training. He also thinks open problems have value for training mathematicians and developing new techniques. So he suggests the community might eventually designate some problems as off limits to automated solvers through social norms, while directing AI toward other problems instead.